SeaEval: leaderboard
Southeast Asian multilingual and cultural reasoning benchmark covering cross-lingual QA, cultural reasoning, safety, and regional language understanding.
Metric: Average Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 72B Instruct | 72.61 | #454 |
| 2 | Qwen 2.5 32B Instruct | 70.56 | #498 |
| 3 | Qwen 2 72B Instruct | 70.53 | #576 |
| 4 | Qwen 2.5 14B Instruct | 69.6 | #647 |
| 5 | Llama 3.1 70B Instruct | 68.37 | #561 |
| 6 | Llama 3.1 70B | 67.78 | #592 |
| 7 | Llama 3.1 70B Cpt Sea Lionv3 Instruct | 66.98 | #548 |
| 8 | Llama 3 70B Instruct | 66.16 | #634 |
| 9 | Qwen 2.5 7B Instruct | 65.94 | #862 |
| 10 | gemma2-9B-cpt-sea-lionv3-instruct | 65.33 | #705 |
| 11 | Gemma 2 9B (IT) | 64.93 | #788 |
| 12 | Qwen 2 7B Instruct | 64.92 | #991 |
| 13 | Llama 3.1 8B CPT SEA LIONv3 Instruct | 64.32 | #810 |
| 14 | SeaLLMs-v3-7B-Chat | 63.37 | #915 |
| 15 | Sailor2-8B-Chat | 59.95 | #782 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seaeval · How It Works · Data refreshed daily, snapshot 2026-10-20.